backend deprecated models check - 2026-10-16

Backend hardcoded model lists compared against https://ai-models.puter.work/ (upstream names = ids + names + aliases, lowercased; a backend model is deprecated only if none of its id/aliases appear upstream). Read-only check, no files modified.

chat/mistral (3)

• mistral-large-4
• mistral-large-2512
• zai-glm-5-3

Notes: Upstream mistral list contains no mistral-large* entries at all (nor mistral-large-latest) and no GLM entries — this contradicts the backend's treatment of Large as flagship, so the endpoint list may be incomplete or key-scoped. Confirm against Mistral directly before removing. zai-glm-5-3 (Z.ai GLM on a mistral provider entry) is not served upstream.

tts/xai (1)

• xai-tts

Notes: Likely a false positive — xai-tts is a pricing key in costs.ts, not a model id, and the xai models endpoint does not list TTS. Verify manually.

No deprecations

chat: claude, openai, gemini, deepseek, moonshot, minimax, xai, alibaba; image: gemini, openai, xai; tts: gemini, openai. (All extracted non-zero model counts.) video/gemini and video/openai: no backend provider exists (ai-video only has byteplus, together), nothing to check.

Verify manually

• chat/zai: 0 flagged, but the zai endpoint lists only base GLM models (11 upstream), so -flash/-flashx/vision variants can't be reliably confirmed either way.